Predicting Travel Intention Using Machine Learning and SHAP-Based Explainability: A Virtual Tourism Approach

Abstract

Virtual tourism is transforming the way users explore and evaluate travel destinations, yet accurately predicting the intention to visit after engaging in virtual experiences remains a challenge. Existing approaches often lack predictive accuracy and interpretability, limiting their application in tourism decision making. To address this, we develop a machine learning framework that integrates explainable artificial intelligence (XAI) to predict the intention to travel after experience with high accuracy and transparency. We implement eXtreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LightGBM), and Multilayer Perceptron (MLP), applying cross-validation for robustness and Optuna-based Bayesian optimization to maximize model performance. To ensure interpretability, we employ SHapley Additive Explanations (SHAP). Our results show that XGBoost outperforms all models, achieving an accuracy of 93.21% and a cross-validation accuracy of 94.08%, validating its
robustness. SHAP analysis reveals that psychological engagement, such as emotional involvement, enjoyment, and immersive flow states, are key drivers of travel intention, with individual SHAP values further elucidating user-specific decision patterns. These findings align with consumer behavior theories, reinforcing the role of psychological and experiential factors in travel decisions. Our study presents a highly accurate, interpretable, and scalable predictive model that advances virtual tourism analytics, providing actionable insights for destination marketing and strategic tourism management. Future research should explore real-time user interactions, adaptive learning techniques, and external variables (social sentiment, economic conditions) to improve predictive accuracy and practical applicability.